Abstract

A large amount of reconstruction work is expected on existing highways due to the fact that highway infrastructures in North America are approaching or have surpassed their service life. The literature of construction engineering and management suggest that urban highway construction projects often overrun in budget and time. Bridges are crucial elements of urban highways, therefore, efficient planning of the construction of bridges is deemed necessary. Bridge construction operations are characterized as equipment-intensive, repetitive, have cyclic nature and involve high uncertainties. Without selecting the best construction method and the optimum number of equipment and crews, projects will take longer and cost more than necessary. The main objectives of this research are to: (1) develop a quantitative method that is capable of obtaining near optimum construction scenarios for bridge construction projects; and (2) obtain these optimum scenarios with an accurate estimate of their objective functions, a high confidence in their optimality and within a short period of time.

The ability of stochastic simulation-based optimization to find near optimum solutions is affected mainly by: (1) the number of candidate solutions generated by the optimization algorithm; and (2) the number of simulation replications required for each candidate solution to achieve a desirable statistical estimate. As a result, a compromise between the accuracy of the estimate of the performance measure index of a candidate solution and the optimality of that candidate solution must be made. Moreover, comparing the performance of different candidate solutions based on the mean values is not accurate because the means of the two objectives (i.e., cost and time) are not always the means of the joint distribution of the two objectives. Finally, the resulting near optimum solutions are not necessarily achievable.

In order to achieve the abovementioned objectives, the following research developments were made: (1) a stochastic simulation-based multi-objective optimization model; (2) a method for incorporating variance reduction techniques into the proposed model; (3) a method to execute the proposed model in parallel computing environment on a single multi-core processor; and (4) a method to apply joint probability to the outcome of the proposed model. The proposed methods showed an average of 84% reduction in the computation time and an average of 18% improvement in the hypervolume indicator over the traditional method when variance reduction techniques are used. Combining variance reduction computing with parallel computing resulted in a time saving of 90%. The use of the joint probability method showed an improvement over the traditional method in the accuracy of selecting the project duration (D) and cost (C) combination that satisfies a certain joint probability. For simulation models with high correlation between the outputs, ΔD and ΔC are not as large as in simulation models with moderate or low correlation, which indicates the existence of a negative relationship between correlation and ΔD and ΔC. In addition, the existence of high correlation permits the reduction of the number of simulation replications required to get a sound estimation of a project, which also indicates the existence of a negative relationship between correlation and the number of replications required.